SOL Trading Β· AI Agents Β· Systems

Two generations.
One terminal.

We build SOL trading infrastructure on Hyperliquid, autonomous AI agent platforms, and crypto data pipelines. Open source. Transparent. No fluff.

Terminal session showing SOL Evo Lab backtest output with graduated strategies
SOL Evo Lab β€” Protocol v3 backtest. 6 graduated strategies, best 53.8% test win, p=0.0066.
6
Graduated strategies
53.8%
Best test win rate
0
API keys for data
1–4
Trades/week cap

Work

Production systems, open research, internal tools. All real, all running.

SOL Evo Lab

Python Hyperliquid Walk-forward Binomial p<0.05

Evolutionary trading lab for SOL perps. Protocol v3: band 311–2170 total trades (1–7/week), graduation requires train winβ‰₯50% + test winβ‰₯50% + nβ‰₯30 + exact binomial p<0.05 vs 33% breakeven null. 6 graduated strategies β€” all volbreak@4h, k 1.16–1.23. Champion 95086ea235: 53.1% test win, p=0.016, +56.7R, maxDD 5.1R. Shorts weak (45.5%) vs longs (57.6%). Forward OOS accrual via daily 01:40 cron on keyless Binance 4h.

Hermes Agent + Miffler

TypeScript Python Docker Telegram Bot API

Self-hosted AI agent platform. Telegram gateway, multi-profile support, skill system, cron jobs, subagent delegation, model provider abstraction (OpenRouter, Z.ai/OAuth, local). Runs on a single VPS with Traefik, Docker, zero external deps for core ops. This conversation is happening on it.

Forensics Lab (Monitor-Only)

Python Pandas Binance 4h Keyless

Post-trade autopsy engine. No killing, no optimization β€” pure observation. Goal: highest win rate, hard 1–4 trades/week cap. RR 1:2, leverage 1–20x. Tracks champion lineage via mt_champions.json. Bridge from lab champions to live paper sim.

Website Builder Stack

Python CLI Geist Geist Mono Railway

End-to-end site generation: build-site CLI (single-file HTML, Geist + Geist Mono, vanilla CSS/JS), Railway deploy, zero external JS deps. Ships production sites in minutes. This site is built with it.

Crypto Free Data Pipeline

Python Binance Bybit CCXT

Keyless funding rates, open interest, liquidations, and candle data from public endpoints. No API keys, no rate limits, no vendor lock-in. Feeds the Evo Lab and Forensics Lab directly.

Live Trading Sim / Paper Engine

Python Hyperliquid SDK Risk gates SSE dashboard

Paper trading simulator with enforced risk gates: RR 1:2, max position sizing, daily loss limits, kill switches. Built to graduate strategies from Forensics β†’ Live without blowing up.

Stack

No hype. Just what survives production.

Python 3.11+ Core / data / trading TypeScript Agent platform, tooling Docker + Traefik Infra, reverse proxy, TLS Hyperliquid SDK Perps execution Pandas / NumPy Research, backtesting CCXT Exchange connectivity Hermes Agent Orchestration, subagents Railway Deploy, HTTPS, logs Geist + Geist Mono Typography Binance public API Keyless 4h candles GitHub Actions CI for agent repo

Timeline

2024
First SOL bot experiments
Early Hyperliquid perps testing, manual discretionary, learning microstructure the expensive way.
Early 2025
Systematic research begins
Built first backtester. Discovered 1–4 week momentum holding period is the crypto sweet spot (vs 3–12mo equities). Vol-adjusted TSMOM + dynamic hold = edge.
Mid 2025
Evo Lab v1 β†’ v3
Protocol v3 hardened: band 311–2170 trades, binomial p<0.05 vs 1/3 null, graduation gates. 6 champions graduated. Refiner mutation steps widened 0.85–1.15.
Late 2025
Hermes Agent adoption
Migrated to self-hosted agent platform. Telegram gateway, skills, subagents, cron orchestration. Dad profile = Tom Helmstaedter (@tomhelmetbot).
2026
Forensics Lab + Website Builder
Monitor-only autopsy engine + end-to-end site generation stack. This site = dogfooding the builder.
Now
Live evolution + selective client work
Paper sim gating, forward OOS accrual, shipping sites for clients via the same stack. No retainers β€” project-based, transparent scope.

About

Two engineers. Different eras. Same stack.

Tom β€” Infrastructure

Decades of systems engineering and infrastructure. The one who asks "what happens when this breaks at 3 AM?" and actually has an answer. Keeps the architecture honest.

  • Systems & infrastructure that doesn't wake you up
  • DevOps pipelines, observability, deployment
  • Code review with teeth
  • Making sure it runs in production, not just locally

rPerpetrader β€” Research & Agents

Quant trading, AI agent orchestration, modern web stacks. Pushes bleeding edge, breaks things, fixes them. Runs the experiments; Dad makes them survivable.

  • SOL perps on Hyperliquid (no KYC, UK VPN)
  • Hermes Agent + Miffler (Telegram gateway, skills, subagents)
  • Evo Lab: walk-forward evolution, binomial gates, forward OOS
  • Forensics Lab: monitor-only autopsies, 1–4 trades/week cap

Contact

We take on select projects: custom websites, AI agent pipelines, crypto data infrastructure, trading tooling. Project-based, transparent scope, no retainers.